The most pressing problem in pairs trading is not the trading rule — it is finding tradeable pairs in the first place. A universe of 5,000 stocks generates roughly 12 million candidate pairs, and running a full cointegration test on all of them is computationally prohibitive. The distance method, popularised by Gatev, Goetzmann, and Rouwenhorst (1999, 2006) and positioned as the first stage of Vidyamurthy's selection pipeline, solves this by assigning every pair a scalar score based on its historical co-movement. Pairs are ranked by score; only the top-ranked candidates advance to statistical testing.
This note develops the distance method end-to-end: how prices are normalized, how the score is computed, and how the formation–trading period split converts the score into a filter with verifiable out-of-sample performance. It also names the two limitations that make the distance method a screening step rather than a trading rule: the absence of economic theory and the convergence trap.
Learning objectives. By the end of this note you should be able to:
- construct a normalized price series and explain why normalization is necessary for cross-sectional comparison;
- compute the sum-of-squared-deviations (SSD) distance between two normalized series;
- describe the formation–trading period architecture and its purpose;
- apply the 2 entry rule and the mean-crossing exit rule;
- identify the two limitations of purely empirical selection and explain why the Vidyamurthy approach addresses them.
In this lesson
-
- The problem: pair selection at scale
-
- Normalized price series
-
- The sum-of-squared-deviations distance
-
- Formation and trading periods
-
- The trading rule
-
- Limitations and the road to cointegration
Before you start
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Statistical Arbitrage syllabus.